Re: [PATCH] It may not be assumed that wake_up(), finish_wait() and co. imply a memory barrier

[Oleg Nesterov <oleg@redhat.com>]

On 04/23, David Howells wrote:
>
> +	complete();
> +	try_to_wake_up();
> +	wake_up();
> +	wake_up_all();
> +	wake_up_bit();
> +	wake_up_interruptible();
> +	wake_up_interruptible_all();
> +	wake_up_interruptible_nr();
> +	wake_up_interruptible_poll();
> +	wake_up_interruptible_sync();
> +	wake_up_interruptible_sync_poll();
> +	wake_up_locked();
> +	wake_up_locked_poll();
> +	wake_up_nr();
> +	wake_up_poll();
> +
> +After waking, and assuming it doesn't take a matching lock, the sleeper may
> +need to interpolate a read or full memory barrier before accessing that state
> +as finish_wait() does not imply a barrier either, and schedule() only implies a
> +barrier on entry.

Well. I am starting to suspect I missed something, but I disagree. Or I just
(this is very possible) misunderstand the above.

finish_wait() doesn't imply a barrier, but why this matters?

And if we don't use prepare_to_wait() and just do

	for (;;) {
		set_current_state(WHATEVER);
		if (!CONDITION)
			schedule();
		break;
	}

we do have mb(), but

> + *
> + * It should not be assumed that this function implies any sort of memory
> + * barrier.
>   */
>  static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync)

unless try_to_wake_up() has a barrier semantics too, this code

	CONDITION = 1;
	wake_up_process(waiter);

is not right, and that mb above can't help.


Could you please give the code example which shows we need a barrier
after finish_wait() ?

I am just trying to understand what I missed.

Oleg.